Re: Streaming and grouping in XSLT 3.0

First question: in your grouping of properties using the rdf:about attribute,
are the groups adjacent? The for-each-group[@group-adjacent] instruction is
fully streamable according to the XSLT 3.0 rules, and the Saxon implementation
is also streamable (though in some circumstances -- I would need to check --
it may build the content of each group in memory before moving on to the
next).

If the groups aren't adjacent, then you can use
xsl:fork/xsl:for-each-group[@group-by], but this isn't really fully streamed
because it will typically collect all the groups in memory.

The next concern is about doing a two-pass transformation where both passes
are streamed. This hasn't really received as much attention as it should in
either the spec or in Saxon. I think it should be possible to pipe the result
of the first transformation directly into the second provided both are written
as free-standing stylesheets (as distinct from a "micro-pipeline" within a
single stylesheet). But I'm not sure I've tested this.

As to the question " how do I even begin analyzing the streamability o this
approach in XSLT 3.0?" I think the answer is [expletive deleted] it and see. First ask
yourself "does it seem intuitively streamable?" in the sense that the Nth
thing in the output can be computed from the Nth thing in the input, plus
perhaps a little bit of memory of earlier things in the input. If the answer
to that is yes, then try writing the code and see if Saxon complains.

Michael Kay
Saxonica

> My use case is as follows -- a triplestore returns streaming RDF/XML
> (omitting namespaces):
>
> <rdf:RDF>
>  <rdf:Description rdf:about="http://dbpedia.org/resource/Copenhagen">
>    <dct:title>Copenhagen</dct:title>
>  </rdf:Description>
>  ...
>  <rdf:Description rdf:about="http://dbpedia.org/resource/Copenhagen">
>    <dbo:country rdf:resource="http://dbpedia.org/resource/Denmark"/>
>  </rdf:Description>
>  ...
> </rdf:RDF>
>
> Since it's streaming, every resource description only contains one
> property. It's not convenient, so during the first pass I group the
> descriptions by subject using [1]:
>
> <rdf:RDF>
>  <rdf:Description rdf:about="http://dbpedia.org/resource/Copenhagen">
>    <dct:title>Copenhagen</dct:title>
>    <dbo:country rdf:resource="http://dbpedia.org/resource/Denmark"/>
>  </rdf:Description>
>  ...
> </rdf:RDF>
>
> Then as the second pass I do the main transformation to XHTML, which
> could produce something like using [2][3]
>
> <fieldset>
>  <label>Title</title>
>  <input type="text" value="Copenhagen"/>
>  <label>Country</title>
>  <input type="text" value="http://dbpedia.org/resource/Denmark"/>
>  ...
> </fieldset>
>
> where fieldsets represent resources, and labels/inputs represent
> properties/values.
>
> Now my question is, how do I even begin analyzing the streamability of
> this approach in XSLT 3.0? I guess my main concern is that such
> grouping would not be streamable, but maybe there are other solutions?
>
> Thanks.
>
> Martynas
>
>
> [1]
https://github.com/AtomGraph/Web-Client/blob/master/src/main/webapp/static/co
m/atomgraph/client/xsl/group-sort-triples.xsl
> [2]
https://github.com/AtomGraph/Web-Client/blob/master/src/main/webapp/static/co
m/atomgraph/client/xsl/bootstrap/2.3.2/layout.xsl
> [3]
https://github.com/AtomGraph/Web-Client/blob/master/src/main/webapp/static/co
m/atomgraph/client/xsl/bootstrap/2.3.2/imports/default.xsl